On 4/24/2015 4:54 AM, Kumaradas Puthussery Krishnadas wrote: > I am creating a SolrCloud with 4 solr instances and 5 zookeeper instances. I > need to make sure that querying is working even when my 3 zookeepers are > down. But it looks like the queries using json transformation based xslt > templates which is not available since the zookeeper ensemble is not > available.
Is it five zookeeper or three? It sounds like you might have five, but three of them are down for some reason. When you have five zookeepers, you can lose two and maintain quorum. If you lose three, then zookeeper doesn't have enough nodes to work properly, and SolrCloud will also stop normal operation. This is a fundamental property of zookeeper. There must be a majority of nodes operational -- more than half. If you have three zookeepers, you can lose only one and still maintain quorum. Some information about zookeeper that isn't directly applicable to your situation but may help explain why zookeeper behaves the way it does: Exactly half of the total nodes is not enough. If you have four zookeepers, two is not enough for quorum, three of them must be operational and able to communicate with each other. This is to prevent split-brain where two clusters are formed that cannot communicate with each other but independently believe that they are the functional cluster. > So is it possible to exclude files (eg: xslt folder) in the conf directory > from being loaded into Zookeeper rather point it to the filesystem so that > querying the solrcloud is not broken. One of the major points of putting the config in zookeeper is to centralize it and have zero reliance on local config files, which may be different on each Solr instance. Consider a cloud with five hundred nodes. A centralized config is the only way to be absolutely certain that every node has the update. You are welcome to file a feature request in Jira for the capability you want, but you may encounter resistance to actually getting it into Solr. If you lose zookeeper quorum, then SolrCloud has no choice other than stopping normal operation, to protect the integrity of the cloud. Any other action could lead to data loss. Zookeeper is a fundamental part of SolrCloud, so if your Zookeeper ensemble is not healthy, neither is SolrCloud. Thanks, Shawn
